Dissertation > Excellent graduate degree dissertation topics show

Study on Particle Detection Technique for Infusion Based on Machine Vision

Author: YangFuGang
Tutor: SunTongJing
School: Shandong University
Course: Control Theory and Control Engineering
Keywords: infusion inspection particle detection Artifictal Immune Algorithm Least Square Support Vector Machines automatic vision inspection
CLC: TP391.41
Type: PhD thesis
Year: 2008
Downloads: 244
Quote: 3
Read: Download Dissertation


Infusion may be mixed with a small amount of particles in the production process. The contamination of particles in infusion solution may come from foreign contamination, such as aluminum scraps, rubber scraps and glass scraps, and they may also be endogenous insoluble solid in raw materials or precipitate after long standing. Particle checking is developed with the intention of detecting impurities in liquid in pharmaceutical containers. According to Chinese Pharmacopoeia, insoluble particle inspection should be conducted in all infusion solutions. Traditional infusion particle checking is labor-intensive and worker-fatigued, and the test results are not stable since there are no unified testing methods and standards. How to check the insoluble particle in infusion effectively has become the bottleneck of automated infusion production line. At present, the study on particle detection of infusion based on machine vision technology is rare. Althouth some foreign research institutions have done some research in the field, particle inspection technology or equipment imported by domestic medical instrument companies can not get satisfactory detection results because of different production environment and pharmaceutical standards. Therefore, the development of online automatic intelligent particle inspection technology suitable for China’s pharmaceutical standards and production environment is significant both theoretically and practically.The key to the particle inspection technology based on machine vision is the effectiveness of image processing algorithms and the high quality of original image acquisition, and the particle detection algorithms is the difficult point and focus of the study. According to the standards of infusion contamination inspection and the resolution of the vision system, two different algorithms are presented in order to detect point target and point extended target respectively, and an ultra-high resolution infusion image acquisition and processing system is also designed to get high-quality original images. The main research is as follows: (1) Particle checking is developed with the intention of detecting impurities in liquid in pharmaceutical containers. This objective has been reached by re-creating a special container analysis condition. The container is made to rotate about its axis in order to agitate the particles and corpuscles in the liquid. Mathematical model of the particles’ centrifugal motion trajectory is established to provide a theory model to utilize the targets’ trajectory characteristics to detect them. Finally, this thesis analyzes the background noise characteristics of infusion image and discusses the evaluation methods of particle detection.(2) The region of interest of the original image is extracted first to reduce computational complexity and improve the efficiency of detection particle. In order to suppress interference from background noise, according to the background noise characteristics of the infusion image, the author presented a background suppression algorithm. Finally, a point extended targets search algorithm is developed, which could separate point extended targets from point targets.(3) In order to avoid getting into local maximum points during the process of optimization and slowing down optimization speed, this thesis improves artificial immune algorithm and developes a new algorithm—an Adaptive Cloning and Suppression Artificial Immune Algorithm. The proposed algorithm, taking into consideration the affinity between antibody and antigen and concentration of antibodies, effectively overcomes the above-mentioned shortcomings in traditional Artificial Immune Algorithm. Meanwhile, this thesis constitutes the mathematical analytical model for the proposed Artificial Immune Algorithm.(4) For point targets, according to their trajectory characteristics in the image sequences, a track optimization algorithm based on the improved Artificial Immune Algorithm is developed. By designing reasonable coding scheme, this thesis constructs hypothetic motion trajectories of these particles, which are regarded as antigens. Then the improved Artificial Immune Algorithm is used to optimize these antigens. Through cloning and mutation, antigens that have maximal affinity and lower concentration are found and are regarded as the possible tracks of particles. Finally, hypothesis testing is used to confirm these possible motion paths and to determine whether particles exist. The experimental results show that by using the algorithm to detect point targets whose sizes are smaller than 50 micron, both the false alarm probablity and detection probability can meet the requirement.(5) To reduce training time and computational complexity of Least Square Support Vector Machines (LSSVM) with kernel of Radial Basis Function, the author developed a novel algorithm for selecting its meta-parameter values based on ideas from principle of artificial immune system. Through analyzing LSSVM parameters on the classification accuracy, it is found that there are many parameters combinations of kernel parameter and regularization parameter that resulted in the same classification accuracy. What’s more, once one of the parameters is fixed and the other is changed in a certain range, their combinations do not affect the classification accuracy. In addition, the coding scheme of many classes to two classes also affects classification results of LSSVM. Therefore, this thesis regards these parameters as antibody genes and designs a reasonable coding scheme for them. Then the improved Artificial Immune Algorithm is employed to search their optimal combination. Experimental results demonstrate that the proposed algorithm greatly enhances parameters optimizing efficiency while keeping the approximately same classification accuracy with some other existent methods such as multi-fold cross-validation and grid-search.(6) An algorithm based on the optimized Least Square Support Vector Machines is proposed to distinguish point extended targets from residual background mass. Through analyzing the infusion image sequences it is found that shape features and gray features of a point extended target have little changes. Therefore, the author first extracts the above-mentioned characteristics of targets from image sequences, and then achieves track association for the same extended target based on optimized Least Square Support Vector Machines. Finally, in accordance with the track feature of the same group of targets associated, the proposed algorithm can distinguish point extended targets from residual background mass. (7) The structure of real-time image acquisition and processing platform for particle inspection is designed in this study. During the design, the author focuses on the solution of ultra-high resolution image acquisition, high-speed image data acquisition and processing technology. Through assigning task to industrial computer and the embedded DSP image-processing platform, real-time contamination of particles detection in infusion could be achieved. Based on the proposed particle detection algorithms, the flow chart of main program of the inspection system and particle detection program are designed. Finally, the images are acquired through this image acquisition experimental platform and processed with high-performance PC, hence the accuracy, speed and repeatability of the detection done by the proposed algorithms is tested.Finally, the summary of the thesis is given, and the future research directions are proposed.

Related Dissertations

  1. Application Study on Electrostatic-Bag Compound Dust Collection for 300MW Thermal Power Unit,X701.2
  2. Researches and Implement of Grain Detection System for Images of Soil Mechanics Experiment,TP391.41
  3. Study on Advanced Prediction Methods of Natural Gas Demand,F407.22
  4. Application and Improvement of Support Vector Machines in Power System Short-Term Load Forecasting,TM715
  5. Study of Method on Non-stationary Time Series Prediction,TP18
  6. The Research on the Finite Difference Time Domain Methods in the Electromagnetic Fields,TM15
  7. Research on Basic Algorithms of Digital Image Processing and Implementation with FPGA,TP391.41
  8. Research on Facial Feature Extraction and Matching Algorithms for Image Retrieval,TP391.41
  9. Research of High Speed Image Pre-processing System Based on FPGA,TP391.41
  10. The Research and Implemention of Image Retrieval Based on User Interested Feature,TP391.41
  11. Research of Image Mosaic Technology,TP391.41
  12. Research on the Classification Based on the Reconstruction of Solder Joint,TP391.41
  13. Tongue Feature Extraction and Research of Fusion Classification,TP391.41
  14. The Fatigue State Recognition of the Driver Based on Eye Detection,TP391.41
  15. Research on Infrared Image Simulation for Aerial Objects and Background,TP391.41
  16. Research on Intelligent Learning-Based Multi-Sensor Target Recognition and Tracking System,TP391.41
  17. Research on Image Compression and Implementation Using TMS320C6713 Based on SPIHT Algorithm,TP391.41
  18. Research on Joint Target Detection for Dual-Sensor Image and System Implementation,TP391.41
  19. Research of Images Enhancing Algorithms on Fog or Backlighting Conditions and Implementation with Hardwares,TP391.41
  20. Video Coding Research Based on Texture Characteristics,TP391.41
  21. Research on Feature Extraction and Classification of Pulse Waveform for Cholecystitis and Nephrotic Syndrome Diagnosis,TP391.41

CLC: > Industrial Technology > Automation technology,computer technology > Computing technology,computer technology > Computer applications > Information processing (information processing) > Pattern Recognition and devices > Image recognition device
© 2012 www.DissertationTopic.Net  Mobile